github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #15184] Installer exits early when nvidia-smi is found even if no nvidia gpu is present #56234

New Issue
Open
opened 2026-04-29 10:27:36 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-29 10:27:36 -05:00
Owner
Copy Link

Originally created by @misieck on GitHub (Mar 31, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15184

What is the issue?

When nvidia-smi is available in the system, the installer thinks its done before trying to discover amd cards and install rocm support.
In fact it appears the installer would skip installalation of cuda as well as that happens later in the script, but im not sure if thats an issue.

nvidia-smi can be installed through the nvidia-utils-xxx package without nvidia hardware present, can be left over from previous hardware installation or there could be both nvidia and amd cards present. So this affects both singe card as well as mixed-card setups.

 # check_gpu nvidia-smi simply checks if nvidia-smi is available to call, i.e. installed in the system

 if check_gpu nvidia-smi; then  
     status "NVIDIA GPU installed."
     exit 0     #        <--------  installer exits 
 fi

 if ! check_gpu lspci nvidia && ! check_gpu lshw nvidia && ! check_gpu lspci amdgpu && ! check_gpu lshw amdgpu; then
     install_success
     warning "No NVIDIA/AMD GPU detected. Ollama will run in CPU-only mode."
     exit 0
 fi
 
 #                      v--------   install rocm support
 if check_gpu lspci amdgpu || check_gpu lshw amdgpu; then
      download_and_extract "https://ollama.com/download" "$OLLAMA_INSTALL_DIR" "ollama-linux-${ARCH}-rocm"   
     install_success
     status "AMD GPU ready."
     exit 0
 fi

To solve this the exit 0 calls should probably be removed in the nvidia and amd check (the latter to allow cuda installation for mixed gpu setups).

Relevant log output

OS

Linux

GPU

AMD

CPU

AMD

Ollama version

0.19.0

Originally created by @misieck on GitHub (Mar 31, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15184 ### What is the issue? When nvidia-smi is available in the system, the installer thinks its done before trying to discover amd cards and install rocm support. In fact it appears the installer would skip installalation of cuda as well as that happens later in the script, but im not sure if thats an issue. nvidia-smi can be installed through the nvidia-utils-xxx package without nvidia hardware present, can be left over from previous hardware installation or there could be both nvidia and amd cards present. So this affects both singe card as well as mixed-card setups. # check_gpu nvidia-smi simply checks if nvidia-smi is available to call, i.e. installed in the system if check_gpu nvidia-smi; then status "NVIDIA GPU installed." exit 0 # <-------- installer exits fi if ! check_gpu lspci nvidia && ! check_gpu lshw nvidia && ! check_gpu lspci amdgpu && ! check_gpu lshw amdgpu; then install_success warning "No NVIDIA/AMD GPU detected. Ollama will run in CPU-only mode." exit 0 fi # v-------- install rocm support if check_gpu lspci amdgpu || check_gpu lshw amdgpu; then download_and_extract "https://ollama.com/download" "$OLLAMA_INSTALL_DIR" "ollama-linux-${ARCH}-rocm" install_success status "AMD GPU ready." exit 0 fi To solve this the exit 0 calls should probably be removed in the nvidia and amd check (the latter to allow cuda installation for mixed gpu setups). ### Relevant log output ```shell ``` ### OS Linux ### GPU AMD ### CPU AMD ### Ollama version 0.19.0
GiteaMirror added the bug label 2026-04-29 10:27:36 -05:00
GiteaMirror commented 2026-04-29 10:27:38 -05:00
Author
Owner
Copy Link

@misieck commented on GitHub (Mar 31, 2026):

For those trying to run AMD cards on systems with NVIDIA installation history, as a workaround uninstall the nvidia-utils-* package before running the installation script.

<!-- gh-comment-id:4165825449 --> @misieck commented on GitHub (Mar 31, 2026): For those trying to run AMD cards on systems with NVIDIA installation history, as a workaround uninstall the nvidia-utils-* package before running the installation script.
GiteaMirror commented 2026-04-29 10:27:39 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 31, 2026):

Or temporarily hide nvidia-smi:

nv=$(command -v nvidia-smi)
sudo mv $nv $nv.tmp
curl -fsSL https://ollama.com/install.sh | sh
sudo mv $nv.tmp $nv
<!-- gh-comment-id:4166053679 --> @rick-github commented on GitHub (Mar 31, 2026): Or temporarily hide `nvidia-smi`: ```shell nv=$(command -v nvidia-smi) sudo mv $nv $nv.tmp curl -fsSL https://ollama.com/install.sh | sh sudo mv $nv.tmp $nv ```
GiteaMirror commented 2026-04-29 10:27:39 -05:00
Author
Owner
Copy Link

@dhirajlochib commented on GitHub (Apr 2, 2026):

Fix submitted in #15225.

The exit 0 after check_gpu nvidia-smi is replaced with a HAS_CUDA flag so the script continues to check for AMD GPUs and install ROCm when needed. This handles both mixed NVIDIA+AMD systems and orphaned nvidia-smi installs.

<!-- gh-comment-id:4179187367 --> @dhirajlochib commented on GitHub (Apr 2, 2026): Fix submitted in #15225. The `exit 0` after `check_gpu nvidia-smi` is replaced with a `HAS_CUDA` flag so the script continues to check for AMD GPUs and install ROCm when needed. This handles both mixed NVIDIA+AMD systems and orphaned nvidia-smi installs.
GiteaMirror commented 2026-04-29 10:27:40 -05:00
Author
Owner
Copy Link

@misieck commented on GitHub (Apr 2, 2026):

Are you sure this will work?
Presence of nvidia-smi alone does not mean cuda is installed does it?
I think check_gpu should not test for existence of nvidia-smi alone, but use it to determine if gpu is found.
Check for cuda should be a separate test, like what is done in the wsl case.

This fix tests for existence of gpus and if its negative, i. e. no gpu installed suppresses warning message if nvidia-smi is found.
It also suppresses installation of cuda if nvidia-smi is installed.

<!-- gh-comment-id:4180624620 --> @misieck commented on GitHub (Apr 2, 2026): Are you sure this will work? Presence of nvidia-smi alone does not mean cuda is installed does it? I think check_gpu should not test for existence of nvidia-smi alone, but use it to determine if gpu is found. Check for cuda should be a separate test, like what is done in the wsl case. This fix tests for existence of gpus and if its negative, i. e. no gpu installed suppresses warning message if nvidia-smi is found. It also suppresses installation of cuda if nvidia-smi is installed.
GiteaMirror commented 2026-04-29 10:27:41 -05:00
Author
Owner
Copy Link

@dhirajlochib commented on GitHub (Apr 2, 2026):

Are you sure this will work? Presence of nvidia-smi alone does not mean cuda is installed does it? I think check_gpu should not test for existence of nvidia-smi alone, but use it to determine if gpu is found. Check for cuda should be a separate test, like what is done in the wsl case.

This fix tests for existence of gpus and if its negative, i. e. no gpu installed suppresses warning message if nvidia-smi is found. It also suppresses installation of cuda if nvidia-smi is installed.

that's a fair point to make sure I implement this correctly are you suggesting something like using nvidia-smi -L to query actual GPU presence, and treating a non-zero exit code as 'no NVIDIA GPU found' even if the binary exists?
Happy to rework the check along those lines before this gets reviewed.

<!-- gh-comment-id:4180684928 --> @dhirajlochib commented on GitHub (Apr 2, 2026): > Are you sure this will work? Presence of nvidia-smi alone does not mean cuda is installed does it? I think check_gpu should not test for existence of nvidia-smi alone, but use it to determine if gpu is found. Check for cuda should be a separate test, like what is done in the wsl case. > > This fix tests for existence of gpus and if its negative, i. e. no gpu installed suppresses warning message if nvidia-smi is found. It also suppresses installation of cuda if nvidia-smi is installed. that's a fair point to make sure I implement this correctly are you suggesting something like using `nvidia-smi -L` to query actual GPU presence, and treating a non-zero exit code as 'no NVIDIA GPU found' even if the binary exists? Happy to rework the check along those lines before this gets reviewed.
GiteaMirror commented 2026-04-29 10:27:44 -05:00
Author
Owner
Copy Link

@PureBlissAK commented on GitHub (Apr 18, 2026):

🤖 Automated Triage & Analysis Report

Issue: #15184
Analyzed: 2026-04-18T18:22:56.674401

Analysis

  • Type: unknown
  • Severity: medium
  • Components: unknown

Implementation Plan

  • Effort: medium
  • Steps:

This issue has been triaged and marked for implementation.

<!-- gh-comment-id:4274310934 --> @PureBlissAK commented on GitHub (Apr 18, 2026): <!-- ollama-issue-orchestrator:v1 issue:15184 --> ## 🤖 Automated Triage & Analysis Report **Issue**: #15184 **Analyzed**: 2026-04-18T18:22:56.674401 ### Analysis - **Type**: unknown - **Severity**: medium - **Components**: unknown ### Implementation Plan - **Effort**: medium - **Steps**: *This issue has been triaged and marked for implementation.*
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#56234
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?